Vertaix (AI&Science Lab at Princeton University)

Our paper on MuAPBEK, a new density functional for quantum chemistry has been published in the Journal of Chemical Physics (JChemPhys or JCP)
🎉

Check it out: pubs.aip.org/aip/jcp/arti...

In our #NewPaper, we showcase the Vendi Score in genomic epidemiology, showing how you can use it to detect new COVID-19 variants before formal identification. Check it out! #Ecology #Genomics #Health

Preprint: arxiv.org/abs/2509.22520

Adji Bousso Dieng, chercheuse en IA : « L’indépendance technologique de l’Afrique est menacée » « Voix d’Afrique » (3/5). L’informaticienne sénégalaise, enseignante à Princeton, appelle à un sursaut de l’Afrique sur la protection numérique des données, une meilleure formation scientifique des pl...

Check out our Adji Bousso Dieng's interview with French newspaper Le Monde on #AI, #Africa, and more:

www.lemonde.fr/afrique/arti...

Are you at the @accelerationc.bsky.social Conference in Toronto? Did you catch Siwoo Lee presenting on "MuAPBEK: An Improved Analytical Kinetic Energy Density Functional For Quantum Chemistry"? 🐯 #ACConference #Vertaix

Check out the paper here: arxiv.org/abs/2505.04559

Highly Accurate and Fast Prediction of MOF Free Energy Via Machine Learning A major bottleneck in translating computational screening of metal-organic frameworks (MOFs) to laboratory synthesis is the uncertain synthetic accessibility of computer-generated MOF designs. While M...

Key ingredients to the solution we developed (continued): a novel sequence representation for MOFs (called MOFSeq), and a pretraining-finetuning strategy for LLM-Prop. 4/4

Check it out! #MatSci #MOFs #LLMs #ML
@chemrxiv.bsky.social link: go.shr.lc/4kIQoWN

Our trained model is also able to accurately select the stable MOF polymorph in a polymorphic family, even when the free energy difference is as low as 0.16 kJ/mol.

Key ingredients to the solution we developed: a new dataset composed of 1 million MOF structures & their properties (MOFMinE) 3/4

The main bottleneck to accelerating the discovery of MOFs is synthesizability. In this paper, we propose a solution to this challenge by developing a holistic machine learning approach to predict MOF free energy and use that to predict MOF synthesizability with 97% accuracy! 2/4

#NewPaper A major development in the design and discovery of metal-organic frameworks (MOFs)! Thread below 👇👇👇 1/4

Our paper introducing Alternators has been accepted at TMLR. Alternators are a new class of generative models that outperform state-space models, such as Mambas, and diffusion models such as ScoreGrad. Check it out!
Paper: arxiv.org/abs/2405.11848
Code: github.com/vertaix/Alternators
#MLSky #Vertaix

Aminah also published her first research paper in CHI 2025. The paper explored how #AI and #AR can be used to facilitate immersive Spanish group conversation practice. ✨ She's joining Apple as a Software Engineer in San Diego, CA. 2/2 #Vertaix #Princeton25

Meet another one of our graduating undergrads, Aminah Aliu🐯🧡

For her thesis, Aminah leveraged the Vendi Score & the Vendiscope to study embedding models on Yoruba datasets.

She was 1 of four Princeton students nominated for the CRA Undergraduate Research Award and won an honorable mention. 1/2

The Princeton Class of 2025 is being hooded today. Meet one of our graduating undergrads, Ian Fernandes! He worked on modeling SEPs using Machine Learning at NASA. He's joining Microsoft. Congratulations, Ian, we are proud of you 🐯🧡 #Vertaix #AI4Science

#NewPaper We revisit Shannon's mutual information, highlight its shortcomings, and propose a novel alternative to it, called the Vendi Information Gain (VIG), using the Vendi Score. #InformationTheory #Diversity
By: Quan Nguyen & @adjiboussodieng.bsky.social
Link to paper: arxiv.org/abs/2505.09007

I couldn't think of a greater way to start the week: "LLM4Mat-Bench: Benchmarking Large Language Models for Materials Property Prediction" (@adjiboussodieng.bsky.social and @jae3goals.bsky.social) is out in #MachineLearningScienceandTechnology 🤩 #EditorsLife
iopscience.iop.org/article/10.1...

Poster promoting an event from the Royal Society of Chemistry, "RSC Desktop Seminar: Americas Emerging Investigators - AI and Machine Learning II" on Monday 14 April 2025, 12pm EST or 5pm UK time

Join us April 14 for the RSC Americas Emerging Investigators webinar on AI/ML in Chemistry! Featuring talks from Jolene Reid (UBC), César de la Fuente (UPenn) @delafuentelab.bsky.social, Adji Bousso Dieng (Princeton) @adjiboussodieng.bsky.social, and Connor Coley (MIT)
rsc.zoom.us/webinar/regi...

Mar 12, 3pm: AC seminar with Adji Bousso Dieng Join us online or in-person for the Acceleration Consortium’s (AC) next seminar of 2025.

Not subscribed to our #substack? You can read the latest edition to learn about our next AC seminar w/ @adjiboussodieng.bsky.social on Mar 12, our summer undergrad research program, our call for abstracts for the annual Accelerate conference + more.

#NewPaper We designed an "algorithmic microscope" & called it the Vendiscope. It is a new companion in the discovery process for scientists across fields and a powerful tool for diagnosing datasets and models for AI researchers. #AlgorithmicMicroscopy

Link to paper: arxiv.org/abs/2502.10828

#NewPaper How do you design generative models for sequence data that are robust to varying noise levels in time? Use the Vendi Score! Check out our new paper introducing the $\alpha$-Alternator, which outperforms popular models such as Mambas on forecasting.

Link to paper: arxiv.org/abs/2502.04593

Vendi-RAG: Adaptively Trading-Off Diversity And Quality Significantly Improves Retrieval Augmented Generation With LLMs Retrieval-augmented generation (RAG) enhances large language models (LLMs) for domain-specific question-answering (QA) tasks by leveraging external knowledge sources. However, traditional RAG systems ...

#NewPaper Introducing Vendi-RAG, a novel RAG pipeline that uses the Vendi Score to balance diversity and relevance in retrieval. This leads to a significant performance boost in multi-hop QA.

Link to paper: arxiv.org/abs/2502.11228
Authors: Mohammad Reza Rezaei and @adjiboussodieng.bsky.social

Congrats to @adjiboussodieng.bsky.social on being named an Early-Career Distinguished Presenter at the MRS Spring Meeting 2025! This honor is meant to highlight exciting new research directions in materials science. 🐯 #ChemScky #Princeton #VendiScoring #MatSci
www.cs.princeton.edu/news/adji-bo...

Behavior-Inspired Neural Networks for Relational Inference From pedestrians to Kuramoto oscillators, interactions between agents govern how a multitude of dynamical systems evolve in space and time. Discovering how these agents relate to each other can improv...

Check out BINNs! Behavior-inspired neural networks for relational inference. To appear at #AISTATS2025.

Paper led by our friends at Princeton MAE 🐯
Link: arxiv.org/abs/2406.14746
#Dynamics #Agents #LatentVariables #AI #Vertaix

Probing out-of-distribution generalization in machine learning for materials - Communications Materials State-of-the-art machine learning models are often tested on their ability to generalize materials deemed ’dissimilar’ to training data, but such definitions frequently rely on heuristics. Here, an an...

Our paper probing out-of-distribution generalization in machine learning for materials discovery is now published at Communications Materials (@NaturePortfolio) #MatSci #AI4Science #Chemsky #Vertaix

📓Link: nature.com/articles/s43...

Our Aminah Aliu won an honorable mention at the 2024-2025 CRA Outstanding Undergraduate Researcher Award 🧡🐯 "This prestigious honor recognizes undergraduate students at North American colleges and universities who demonstrate exceptional potential in computing research." #CRAAwards #Vertaix

Nice to meet you @andreniyongabo.bsky.social , great work!
👋 @adjiboussodieng.bsky.social